AbstractsGeography &GIS

Airborne Laser scanning (ALS) has emerged as an important tool for providing costeffective characterisation of the 3D structure of forests over large areas. As data resolution is often inversely proportional to coverage area, laser scanning from alternative platforms has been a recent subject of investigation. This thesis advances this exploration by investigating the use of Unmanned Aerial Vehicles (UAVs) as a laser scanning platform (UAVLS) for forest inventory purposes. The design of a small laser scanning system consisting of an automotive laser scanner, a Micro-Electro-Mechanical Systems based Inertial Measurement Unit (IMU), a dual frequency Global Positioning System (GPS) receiver and a downward pointing video camera for use on-board an Oktokopter multirotor platform is described. A novel algorithm was developed for the direct georeferencing of laser returns utilising a vision aided GPS-IMU sigma-point Kalman smoother. Evaluating improvements due to the inclusion of vision, both stochastically and in practice, it is demonstrated that an accuracy similar to modern ALS systems and adequate for forest inventory measurements can be achieved (34 cm horizontal, 14 cm vertical RMSE). Two 4 year old Eucalyptus plantations in south east Tasmania were selected as the primary study area in order to assess the utility of the UAVLS system to map and assess change in key inventory metrics. Analysis of the point clouds captured with different flying parameters indicated that the flying height should be restricted to less than 50 m above ground level and scan angle restricted to ±30. A survey method within these restraints and utilising overlapping transects was designed to provide cost-effective and repeatable observations of the 3D structure of the plot sized areas (500 m2). It was found that the maximum deviations of plot level descriptive statistics captured in repeat multiple flights were less than 3%. Investigating the accuracy and repeatability of individual tree level metrics derived from the high density UAVLS point clouds (up to 300 points/m2) using five different automatic tree detection and delineation methods highlighted that increased data resolution provided more detail in the characterisation of individual trees. The best performing method, which utilised both the CHM and the point cloud, resulted in 98% of trees being repeatedly and correctly delineated from the point cloud. Tree height (absolute mean deviation of 0.35 m), location (0.48 m), crown area (3.3 m2) and canopy closure (2.3%) extracted from the delineated tree segments were observed with higher repeatability and better efficiency than that currently achieved using modern field techniques. Subsequent analysis of change following the application of sequential silvicultural treatments showed that UAVLS is capable of detecting pruning rates of between 96 and 125% of the true pruning rate. This thesis demonstrates that UAVLS offers unprecedented temporal and spatial resolution, enabling the determination of highly…